Jim Keller on Open AI Architecture, Chiplets, and RISC-V

Jim Keller's Engineering Philosophy

• 💡 Simplicity and Modularity: Jim Keller emphasizes that complicated systems cannot be fixed; instead, complex systems must be built from simpler, modular components with clear abstraction layers.
• 🧠 Innovation Approach: His approach to innovation has evolved from processor-centric roles to AI-focused, distributed, and data-centric systems, requiring constant re-engineering at component levels.

Evolving Compute Architectures

• 🚀 Data Movement as Key: Performance in modern AI systems is increasingly dependent on efficient data movement across various scales (on-die, chiplets, systems), rather than just raw computational power.
• 🧩 Heterogeneous Computing: The future involves CPUs and AI processors collaborating in novel, flexible ways, moving beyond traditional host CPU-driven GPU models.
• ✅ Software Stack Simplification: The goal is to build layered software stacks that allow AI model programmers to simply describe their needs, reducing the complexity of managing multiple software layers.

Chiplets and Open Standards

• 🏗️ Chiplet-Based Design: Tenstor refactored its chip plans to be chiplet-based, enabling the construction of highly complex systems from simpler, modular components.
• 🌐 Open Chiplet Atlas (OCA): Tenstor developed its own Open Chiplet Atlas (OCA) framework to ensure chiplet compatibility and rapid iteration, prioritizing speed over committee-driven standards.
• 💡 RISC-V and Tensic Architectures: RISC-V is highlighted as an open CPU architecture platform for innovation, and Tenstor aims for Tensic to be the open AI architecture, allowing for modification and experimentation.

Democratizing AI Innovation

• 🌍 Sovereign Compute Platforms: Many regions seek to own their technology stack, and Tenstor enables this by offering IP, chiplets, and full-stack solutions, fostering local technology development.
• 🤝 Partnership and Licensing: Tenstor licenses its AI processor IP to companies like LG and Boss, providing an accessible on-ramp for partners to build their own silicon dreams without doing everything from scratch.

Mentoring Future Architects

• 🌱 Focus on Openness and Impact: Jim advises future architects to work on open projects that allow for modification and collaboration, ideally in small companies where they can have a big impact.
• 🛠️ Hands-on and Fast-Paced: Emphasizes being hands-on, writing code daily, and working in environments that prioritize velocity and leverage AI for code generation and refactoring.